In Illusory Flash (IF) experiments, congruent multisensory presentation has no effect on the mean estimate of the number of events, but decreases the variance in comparison with unisensory presentation. In contrast, congruent multisensory presentation in other Temporal Numerosity Judgement (TNJ) tasks affects the mean estimate (i.e., it often results in a reduction in underestimation) and increases the variance. In three experiments, we investigated the differences between both paradigms as possible causes of this discrepancy: the presence or absence of incongruent stimuli (Experiment 1), the instruction to the observer to either count flashes, beeps or multisensory events (Experiment 2), and the range of pulses presented (Experiment 3). We found significant differences between the mean numerosity estimate of multisensory and unisensory series in Experiment 3, but not in 1 and 2. This suggests that the difference in the range of pulses presented in IF (1-3 pulses) and TNJ (1-10 pulses) is the primary cause of the discrepancy. In the discussion we propose that this result may be explained by the use of two different strategies and their susceptibility to multisensory presentation. For small pulse numbers, observers can accurately count both unisensory and multisensory pulses. For larger numbers, observers can no longer count but will estimate the number based on the pulse series duration which is improved for multisensory stimuli.